2026-07-21 Digest

Lead: Cost‑impacting change

• Z.ai: GLM 5.2 – Prompt price fell from $1.12/1M to $0.94/1M; completion price fell from $3.52/1M to $2.94/1M. Total cost per million tokens dropped from $4.64 to $3.88 (‑$0.76/1M, ~16% lower). Who should care: Teams running high‑volume GLM 5.2 workloads see noticeable inference‑cost savings.

Other price changes

• Tencent: Hy3 – Prompt $0.20/1M → $0.14/1M; completion $0.80/1M → $0.58/1M; total $1.00 → $0.72/1M (‑$0.28/1M).
• MoonshotAI: Kimi K2.7 Code – Prompt $0.85/1M → $0.82/1M; completion $3.80/1M → $3.75/1M; total $4.65 → $4.57/1M (‑$0.08/1M).
• DeepSeek: DeepSeek V4 Flash – Prompt $0.098/1M → $0.094/1M; completion $0.196/1M → $0.188/1M; total $0.294 → $0.281/1M (‑$0.013/1M).
• Google: Gemma 4 31B – Prompt $0.22/1M → $0.12/1M; completion $0.55/1M → $0.37/1M; total $0.77 → $0.49/1M (‑$0.28/1M).
• NVIDIA: Nemotron 3 Super – Prompt $0.085/1M → $0.08/1M; completion $0.40/1M → $0.45/1M; total $0.485 → $0.53/1M (+$0.045/1M, slight increase).
• Qwen: Qwen3 Next 80B A3B Instruct – Prompt $0.10/1M → $0.098/1M; completion $1.10/1M → $0.78/1M; total $1.20 → $0.878/1M (‑$0.323/1M).

Added model

• Meituan: LongCat 2.0 – Prompt $0.30/1M, completion $1.20/1M → total $1.50/1M. Context 1,048,756 tokens. Who should care: Users needing very long context (≈1M tokens) at a moderate price point.

Removed model

• Tencent: Hy3 (free) – Zero‑cost model with 262,144‑token context removed. Who should care: Anyone relying on the free Hy3 tier must migrate to a paid alternative or another free model.

---

Originally published at The Token Ledger. Subscribe for the daily digest.